EU-LIFE

Building and promoting Excellence in Life Sciences in Europe
Building and promoting Excellence in Life Sciences in Europe

Master Student - Convolutional neural networks for localization of high-grade prostate cancer using quantitative MRI

Description

Your function within the department

Research question 
Investigate the feasibility of applying a recent class of CNNs (fully convolutional) to the segmentation of high Gleason grade regions in prostate cancer using quantitative MRI

Project description 
The Gleason score (GS), a well-validated prognostic factor, estimates the aggressiveness of prostate cancer. The GS is usually determined during the diagnostic phase from random biopsies, but it often represents an under- or overestimation compared to the GS determined on whole-mount section histology. As multi-parametric (mp-) MRI can be used to determine the GS, targeted biopsies based on mp-MRI are being considered as an alternative to systematic biopsies. To improve the accuracy of MR-guided targeted biopsies, non-invasive methods need to be developed that give spatial information about the presence of high-grade GP within heterogeneous prostate tumours. We therefore aim to assess the accuracy of a combination of quantitative mp-MRI (qMRI) techniques (T2 map, diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) MRI) for spatial mapping of high-grade GP in a prospective three-centre study.
Traditional machine learning-based techniques have been proposed that perform the segmentation of GS using texture features extracted from mp-MR. More recently, deep learning (DL) approaches have shown promising results in medical image segmentation tasks and are now receiving increasingly more attention in the community. In particular, a first attempt has been made that uses convolutional neural networks (CNN) to distinguish between different GS. However, there was no proper validation with the true GS regions as determined using histopathology.
In this study, we would like to investigate the feasibility of applying a recent class of CNNs (fully convolutional) to the segmentation of different GS regions in our dataset.

Your profile

  • Basic knowledge on deep/machine learning techniques
  • Basic programming skills

Your career opportunities and terms of employment

Start date: May 2018
Expected final date: After 6 months
Work load (weeks): 24 weeks
 

Interested?

For information, contact:

Rita Simões, tel: 020 512 2262, r.simoes@nki.nl
Ghazaleh Ghobadi, tel: 020 512 1744, 
  

How to apply

Apply now

Host Institute